Automated crystal identification achieved via watershed segmentation

Abstract

A nuclear imaging system includes a crystal identification system which receives a flood image which includes a plurality of peaks, each peak responsive to radiation detected by a corresponding scintillator crystal. A crystal identification processor partitions the flood image into a plurality of candidate regions with a watershed segmentator implementing a watershed algorithm. The candidate regions are linked in an adjacency graph, and then classified as background or relevant, where relevant regions contain a peak within the watershed lines. The regions are then assigned to a crystal according to an objective function and an assignability score. A calibration processor maps the peaks to a rectangular grid.

The invention claimed is: 
     
       1. A method for crystal identification in nuclear imaging systems, comprising:
 generating a flood image which includes a plurality of peaks, each peak responsive to received radiation and corresponding to a scintillator crystal; 
 identifying candidate regions in the flood image with a watershed segmentator implementing a watershed transform; 
 assigning crystal for each candidate region; and, 
 aligning the identified candidate regions of the flood image according to a rectangular grid. 
 
     
     
       2. The method according to  claim 1 , wherein the identifying step includes:
 generating watershed lines according to minima within the flood image to form candidate regions; and 
 generating an adjacency graph with an adjacency graph generator to link the candidate regions within the flood image. 
 
     
     
       3. The method according to  claim 1 , wherein the assigning step includes:
 assigning candidate regions with an intensity below a defined threshold as a background region. 
 
     
     
       4. The method according to  claim 3 , wherein the assigning step includes:
 assigning non-background candidate regions to individual crystals. 
 
     
     
       5. The method according to  claim 3 , wherein the assigning step includes:
 calculating an assignability score for each region; 
 selecting the highest assignability scored region; 
 assigning the selected region to a grid location with crystal coordinates; and 
 assigning adjacent regions according to a selected objective function. 
 
     
     
       6. The method according to  claim 5 , wherein the assigning step includes:
 for regions with low assignability score, selecting a combination of crystal coordinates according to the objective function. 
 
     
     
       7. The method according to  claim 3 , wherein the assigning step includes:
 renumbering the crystal coordinates. 
 
     
     
       8. A non-transitory computer readable medium carrying software for controlling one or more processors to perform the method of  claim 1 . 
     
     
       9. A crystal identification system, comprising:
 an image memory which receives a flood image which includes a plurality of peaks, each peak responsive to radiation detected by a corresponding scintillator crystal; 
 a crystal identification processor configured to:
 identify candidate regions with an identification processor implementing a watershed transform; 
 
 assign a crystal for each candidate region; 
 align the candidate regions according to a rectangular grid for the flood image. 
 
     
     
       10. The crystal identification system according to  claim 9 , wherein the crystal identification processor is further programmed to identify the candidate regions by:
 generating watershed lines with a watershed segmentator according to minima within the flood image; and 
 generating an adjacency graph with an adjacency graph generator to link the candidate regions within the flood image. 
 
     
     
       11. The crystal identification system according to  claim 9 , wherein the crystal identification processor is further programmed to assign the candidate regions by:
 assigning candidate regions having an intensity below a defined threshold as a background region. 
 
     
     
       12. The crystal identification system according to  claim 11 , wherein the crystal identification processor is further programmed to assign the candidate regions by:
 assigning non-background candidate regions to individual crystals. 
 
     
     
       13. The crystal identification system according to  claim 12 , wherein the crystal identification processor is further programmed to assign the candidate regions by:
 calculating an assignability score for each region; 
 selecting the highest assignability scored region; 
 assigning the selected region to a grid location with crystal coordinates; and 
 assigning adjacent regions according to a selected objective function. 
 
     
     
       14. The crystal identification system according to  claim 13 , wherein the crystal identification processor is further programmed to assign the candidate regions by:
 renumbering the crystal coordinates. 
 
     
     
       15. The crystal identification system according to  claim 12 , wherein the crystal identification processor is further programmed to assign the candidate regions by:
 for regions with a low assignability score, selecting an optimal combination according to the objective function. 
 
     
     
       16. An apparatus for crystal identification in nuclear imaging systems with a crystal identification processor configured to:
 generate a flood image; 
 generating watershed lines along local intensity minima within the flood image to create candidate regions; 
 partition the flood image into a plurality of candidate regions; 
 identify maxima within each candidate region; 
 assign the candidate regions to corresponding crystals; and 
 assign the identified maxima to a crystal having crystal coordinates. 
 
     
     
       17. The apparatus according to  claim 16 , the crystal identification processor further configured to:
 segment the flood image into candidate regions using a watershed algorithm; 
 generate an adjacency graph to link candidate regions); 
 locate a peak within each candidate region; 
 label each peak with x/y coordinates according to an assignability score and a defined optimality objective function; 
 map the adjacency graph to a rectangular template with a calibration processor; and 
 display the mapped adjacency graph to a user via a graphical user interface.